[X86] Add atomic vector tests for unaligned >1 sizes. #120387

jofrn · 2024-12-18T08:41:44Z

Unaligned atomic vectors with size >1 are lowered to calls.
Adding their tests separately here.

Stack:

⚠️ Part of a stack created by spr. Do not merge manually using the UI - doing so may have unexpected results.

llvmbot · 2024-12-18T08:42:21Z

@llvm/pr-subscribers-backend-x86

Author: None (jofrn)

Changes

Atomic vectors with size >1 are lowered to calls. Adding
their tests separately here.

Stack:

#120387 ⬅
#120386
#120385
#120384

⚠️ Part of a stack created by spr. Do not merge manually using the UI - doing so may have unexpected results.

Full diff: https://github.com/llvm/llvm-project/pull/120387.diff

1 Files Affected:

(modified) llvm/test/CodeGen/X86/atomic-load-store.ll (+223)

diff --git a/llvm/test/CodeGen/X86/atomic-load-store.ll b/llvm/test/CodeGen/X86/atomic-load-store.ll
index 2bde0d2ffd06ad..435e58bef6642d 100644
--- a/llvm/test/CodeGen/X86/atomic-load-store.ll
+++ b/llvm/test/CodeGen/X86/atomic-load-store.ll
@@ -110,3 +110,226 @@ define <1 x bfloat> @atomic_vec1_bfloat(ptr %x) {
   ret <1 x bfloat> %ret
 }
 
+define <1 x i64> @atomic_vec1_i64(ptr %x) {
+; CHECK-LABEL: atomic_vec1_i64:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $8, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movq (%rsp), %rax
+; CHECK-NEXT:    popq %rcx
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec1_i64:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    pushq %rax
+; CHECK0-NEXT:    .cfi_def_cfa_offset 16
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $8, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movq (%rsp), %rax
+; CHECK0-NEXT:    popq %rcx
+; CHECK0-NEXT:    retq
+  %ret = load atomic <1 x i64>, ptr %x acquire, align 4
+  ret <1 x i64> %ret
+}
+
+define <1 x double> @atomic_vec1_double(ptr %x) {
+; CHECK-LABEL: atomic_vec1_double:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $8, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec1_double:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    pushq %rax
+; CHECK0-NEXT:    .cfi_def_cfa_offset 16
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $8, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK0-NEXT:    popq %rax
+; CHECK0-NEXT:    retq
+  %ret = load atomic <1 x double>, ptr %x acquire, align 4
+  ret <1 x double> %ret
+}
+
+define <2 x i32> @atomic_vec2_i32(ptr %x) {
+; CHECK-LABEL: atomic_vec2_i32:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $8, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK-NEXT:    popq %rax
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec2_i32:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    pushq %rax
+; CHECK0-NEXT:    .cfi_def_cfa_offset 16
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $8, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movq {{.*#+}} xmm0 = mem[0],zero
+; CHECK0-NEXT:    popq %rax
+; CHECK0-NEXT:    retq
+  %ret = load atomic <2 x i32>, ptr %x acquire, align 4
+  ret <2 x i32> %ret
+}
+
+define <4 x float> @atomic_vec4_float(ptr %x) {
+; CHECK-LABEL: atomic_vec4_float:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    subq $24, %rsp
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $16, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movaps (%rsp), %xmm0
+; CHECK-NEXT:    addq $24, %rsp
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec4_float:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    subq $24, %rsp
+; CHECK0-NEXT:    .cfi_def_cfa_offset 32
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $16, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movaps (%rsp), %xmm0
+; CHECK0-NEXT:    addq $24, %rsp
+; CHECK0-NEXT:    retq
+  %ret = load atomic <4 x float>, ptr %x acquire, align 4
+  ret <4 x float> %ret
+}
+
+define <8 x double> @atomic_vec8_double(ptr %x) {
+; CHECK-LABEL: atomic_vec8_double:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    subq $72, %rsp
+; CHECK-NEXT:    .cfi_def_cfa_offset 80
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $64, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movaps (%rsp), %xmm0
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm1
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm2
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm3
+; CHECK-NEXT:    addq $72, %rsp
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec8_double:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    subq $72, %rsp
+; CHECK0-NEXT:    .cfi_def_cfa_offset 80
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $64, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movapd (%rsp), %xmm0
+; CHECK0-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm1
+; CHECK0-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm2
+; CHECK0-NEXT:    movapd {{[0-9]+}}(%rsp), %xmm3
+; CHECK0-NEXT:    addq $72, %rsp
+; CHECK0-NEXT:    retq
+  %ret = load atomic <8 x double>, ptr %x acquire, align 4
+  ret <8 x double> %ret
+}
+
+define <16 x bfloat> @atomic_vec16_bfloat(ptr %x) {
+; CHECK-LABEL: atomic_vec16_bfloat:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    subq $40, %rsp
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $32, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movaps (%rsp), %xmm0
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm1
+; CHECK-NEXT:    addq $40, %rsp
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec16_bfloat:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    subq $40, %rsp
+; CHECK0-NEXT:    .cfi_def_cfa_offset 48
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $32, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movaps (%rsp), %xmm0
+; CHECK0-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm1
+; CHECK0-NEXT:    addq $40, %rsp
+; CHECK0-NEXT:    retq
+  %ret = load atomic <16 x bfloat>, ptr %x acquire, align 4
+  ret <16 x bfloat> %ret
+}
+
+define <32 x half> @atomic_vec32_half(ptr %x) {
+; CHECK-LABEL: atomic_vec32_half:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    subq $72, %rsp
+; CHECK-NEXT:    .cfi_def_cfa_offset 80
+; CHECK-NEXT:    movq %rdi, %rsi
+; CHECK-NEXT:    movq %rsp, %rdx
+; CHECK-NEXT:    movl $64, %edi
+; CHECK-NEXT:    movl $2, %ecx
+; CHECK-NEXT:    callq ___atomic_load
+; CHECK-NEXT:    movaps (%rsp), %xmm0
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm1
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm2
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm3
+; CHECK-NEXT:    addq $72, %rsp
+; CHECK-NEXT:    retq
+;
+; CHECK0-LABEL: atomic_vec32_half:
+; CHECK0:       ## %bb.0:
+; CHECK0-NEXT:    subq $72, %rsp
+; CHECK0-NEXT:    .cfi_def_cfa_offset 80
+; CHECK0-NEXT:    movq %rdi, %rsi
+; CHECK0-NEXT:    movl $64, %edi
+; CHECK0-NEXT:    movq %rsp, %rdx
+; CHECK0-NEXT:    movl $2, %ecx
+; CHECK0-NEXT:    callq ___atomic_load
+; CHECK0-NEXT:    movaps (%rsp), %xmm0
+; CHECK0-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm1
+; CHECK0-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm2
+; CHECK0-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm3
+; CHECK0-NEXT:    addq $72, %rsp
+; CHECK0-NEXT:    retq
+  %ret = load atomic <32 x half>, ptr %x acquire, align 4
+  ret <32 x half> %ret
+}

RKSimon · 2024-12-18T10:07:53Z

llvm/test/CodeGen/X86/atomic-load-store.ll

@@ -110,3 +110,226 @@ define <1 x bfloat> @atomic_vec1_bfloat(ptr %x) {
  ret <1 x bfloat> %ret
 }

+define <1 x i64> @atomic_vec1_i64(ptr %x) {


add nounwind to get rid of cfi noise

RKSimon

LGTM - cheers

jyknight · 2024-12-18T23:11:51Z

Atomic vectors with size >1 are lowered to calls.

That's not true; they're only lowered to calls when the alignment is not known to be sufficient (e.g. <2 x i32> must have align 8, not align 4).

arsenm · 2024-12-19T01:31:19Z

llvm/test/CodeGen/X86/atomic-load-store.ll

+; CHECK0-NEXT:    movq (%rsp), %rax
+; CHECK0-NEXT:    popq %rcx
+; CHECK0-NEXT:    retq
+  %ret = load atomic <1 x ptr>, ptr %x acquire, align 4


Should test the naturally aligned case, this under-aligned one is a separate test

Added that one here: https://github.com/llvm/llvm-project/pull/120385/files.

Probably should just add these kinds of cases with that PR. The set of strange under-aligned cases happen to work now, but I'm not sure that's enough reason to separately push them

llvm/test/CodeGen/X86/atomic-load-store.ll

arsenm · 2024-12-19T04:05:06Z

llvm/test/CodeGen/X86/atomic-load-store.ll

+; CHECK0-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
+; CHECK0-NEXT:    popq %rax
+; CHECK0-NEXT:    retq
+  %ret = load atomic <1 x double>, ptr %x acquire, align 4


these are all under aligned

Added this one here: https://github.com/llvm/llvm-project/pull/120386/files.

arsenm

Tests are fine but I'd just merge these in with the support commits

Unaligned atomic vectors with size >1 are lowered to calls. Adding their tests separately here. commit-id:a06a5cc6

This was referenced Dec 18, 2024

[SelectionDAG] Legalize <1 x T> vector types for atomic load #120385

Open

IR/Verifier: Allow vector type in atomic load and store #120384

Open

[X86] Manage atomic load of fp -> int promotion in DAG #120386

Open

llvmbot added the backend:X86 label Dec 18, 2024

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 441e75a to 3773e66 Compare December 18, 2024 08:54

RKSimon reviewed Dec 18, 2024

View reviewed changes

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 3773e66 to 813fffe Compare December 18, 2024 11:45

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from 141279f to 70bb5b9 Compare December 18, 2024 11:45

jofrn requested review from arsenm, shiltian and Pierre-vh December 18, 2024 11:48

jofrn mentioned this pull request Dec 18, 2024

[X86] Add atomic vector tests for >1 sizes. #120316

Open

RKSimon approved these changes Dec 18, 2024

View reviewed changes

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from 70bb5b9 to dac7f1e Compare December 18, 2024 19:11

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch 2 times, most recently from 05a76cf to bb71e93 Compare December 18, 2024 20:47

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from dac7f1e to df5e28c Compare December 18, 2024 20:47

arsenm reviewed Dec 19, 2024

View reviewed changes

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from bb71e93 to 5e8da05 Compare December 19, 2024 02:29

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch 2 times, most recently from 255a011 to e3dd939 Compare December 19, 2024 02:31

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 5e8da05 to e71ac05 Compare December 19, 2024 02:31

arsenm reviewed Dec 19, 2024

View reviewed changes

jofrn changed the title ~~[X86] Add atomic vector tests for >1 sizes.~~ [X86] Add atomic vector tests for unaligned >1 sizes. Dec 19, 2024

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from e3dd939 to b336c25 Compare December 19, 2024 13:16

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch 2 times, most recently from 0564ecb to 454b8e6 Compare December 19, 2024 16:01

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from b336c25 to 7ef2576 Compare December 19, 2024 16:01

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from 19fb0ba to d2b0634 Compare January 7, 2025 15:31

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from d2b0634 to 9a7ee21 Compare January 15, 2025 11:52

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch 2 times, most recently from 84a1672 to 6d84ebe Compare January 15, 2025 13:26

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from b2541fd to 846ab2e Compare January 21, 2025 16:58

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 6d84ebe to 647a59b Compare January 21, 2025 16:58

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from 846ab2e to 4698589 Compare January 21, 2025 17:13

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 647a59b to b0364ee Compare January 21, 2025 17:13

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from 4698589 to e261f40 Compare January 21, 2025 17:44

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from b0364ee to 6667568 Compare January 21, 2025 17:44

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from e261f40 to 6f50ac7 Compare January 21, 2025 17:50

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 6667568 to 2af4d7c Compare January 21, 2025 17:50

arsenm approved these changes Jan 22, 2025

View reviewed changes

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 2af4d7c to 7562f64 Compare January 22, 2025 11:05

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch 2 times, most recently from 8137157 to a495eaa Compare January 22, 2025 11:43

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 7562f64 to bc9ddb8 Compare January 22, 2025 11:43

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from a495eaa to ab42ec1 Compare January 22, 2025 17:38

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch 2 times, most recently from 657d3dc to 86d8d1c Compare January 22, 2025 17:47

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from ab42ec1 to 6ac3c17 Compare January 22, 2025 17:47

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 86d8d1c to 3c692ed Compare January 22, 2025 18:19

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from 6ac3c17 to 03a726d Compare January 22, 2025 18:19

jofrn changed the base branch from users/jofrn/spr/main/f9d761c5 to main February 2, 2025 20:25

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 3c692ed to 05e7afd Compare February 2, 2025 20:25

jofrn mentioned this pull request Feb 2, 2025

[SelectionDAG][X86] Remove unused elements from atomic vector. #125432

Open

jofrn changed the base branch from main to users/jofrn/spr/main/f9d761c5 February 2, 2025 20:25

[X86] Add atomic vector tests for unaligned >1 sizes.

03400f8

Unaligned atomic vectors with size >1 are lowered to calls. Adding their tests separately here. commit-id:a06a5cc6

jofrn force-pushed the users/jofrn/spr/main/a06a5cc6 branch from 05e7afd to 03400f8 Compare March 3, 2025 23:26

jofrn force-pushed the users/jofrn/spr/main/f9d761c5 branch from f096c88 to edd2af8 Compare March 3, 2025 23:27

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[X86] Add atomic vector tests for unaligned >1 sizes. #120387

[X86] Add atomic vector tests for unaligned >1 sizes. #120387

jofrn commented Dec 18, 2024 •

edited

Loading

llvmbot commented Dec 18, 2024

RKSimon Dec 18, 2024

RKSimon left a comment

jyknight commented Dec 18, 2024

arsenm Dec 19, 2024

jofrn Dec 20, 2024

arsenm Jan 6, 2025

arsenm Dec 19, 2024

jofrn Dec 20, 2024

arsenm left a comment

[X86] Add atomic vector tests for unaligned >1 sizes. #120387

Are you sure you want to change the base?

[X86] Add atomic vector tests for unaligned >1 sizes. #120387

Conversation

jofrn commented Dec 18, 2024 • edited Loading

llvmbot commented Dec 18, 2024

RKSimon Dec 18, 2024

Choose a reason for hiding this comment

RKSimon left a comment

Choose a reason for hiding this comment

jyknight commented Dec 18, 2024

arsenm Dec 19, 2024

Choose a reason for hiding this comment

jofrn Dec 20, 2024

Choose a reason for hiding this comment

arsenm Jan 6, 2025

Choose a reason for hiding this comment

arsenm Dec 19, 2024

Choose a reason for hiding this comment

jofrn Dec 20, 2024

Choose a reason for hiding this comment

arsenm left a comment

Choose a reason for hiding this comment

jofrn commented Dec 18, 2024 •

edited

Loading